|
|
|
|
|
|
|
|
|
|
xen-users
Re: [Xen-users] Xen and PPC and ARM
> New crazy questions.
Good :-)
> Is it possible to make PCI passthrough without
> domU stopping?
There were some patches posted recently to allow PCI devices to be hot-plugged
and hot-unplugged into a privileged domU. I *think* they even allowed you to
hot-unplug PCI devices and then migrate the domain (you can't ordinarily
migrate a domain which has PCI devices passed through to it).
> I know that I can hide devices from dom0
> without reboot, but what about hot PCI devices passing?
> Is it possible?
The patches that make it possible are already in xen-unstable, I believe.
See: http://xenbits.xensource.com/xen-unstable.hg?rev/3f407392da49
This looks like you can now hot-add and -remove PCI devices from PV guests. I
think other changesets also made it possible for HVM guests if you have VT-d
hardware.
I've not tried this and I don't know if there are any special requirements in
the guest kernel, in order for it to handle this.
> Is it possible (at least theoretically) to change dom0 into domU and
> some domU into dom0 without rebooting the system?
> AFAIU, to do this we should pass all dom0 devices to
> another domain and after that make reattaching of the
> running domUs devices from current dom0 to the "new dom0".
Ah, well that's an interesting question.
There are a lot of devices that would need passing over and not all of them
would be PCI devices. So you couldn't do it just with hotplug. In fact,
simple PCI hotplug isn't enough on its own, since a domain that's had a PCI
device passed through to it still has to occasionally get help from dom0,
which is more privileged.
Pretty much anything is possible if you put your mind to it, but I don't think
I'm qualified to say whether doing this would be worthwhile in terms of
effort (or efficient in terms of performance).
You'd not only need the new dom0 to take over the low level devices in the
system, you'd also need to keep the management state (Xend's state,
xenstored's state) in sync between the old dom0 and the new dom0 so that it
could go on managing the domUs correctly.
I think it would be quite tricky to get this working, though!
> Or more easy task.
>
> Presume,
> we have maid lightweight dom0
> that uses initrd only (he also can use external devices
> when booting new domains).
> And than we have maid lightweight domU (DOMAIN1)
> that uses initrd only. It does not use network and disks.
>
> At some point we have these two domains running (dom0 and DOMAIN1).
> None of them uses disk and network.
>
> Can we make change DOMAIN1 into dom0 (and former dom0 into say DOMAIN2)?
Again, the problem is in all the low level transfers of control. It's still
going to be a problem here.
Again, it's probably doable with enough coding skill and low level knowledge
but would be very tricky.
> Or it's crazy idea ant it's better not to think about it?
It's not a crazy idea - you might want (for instance) to have a "failsafe"
dom0 that kicks in if the primary dom0 dies. However, nobody (that I know
of) has done this yet and that's at least partly because it's a hard
problem ;-) Definitely interesting to think about, though!
Cheers,
Mark
--
Push Me Pull You - Distributed SCM tool (http://www.cl.cam.ac.uk/~maw48/pmpu/)
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
<Prev in Thread] |
Current Thread |
[Next in Thread> |
- Re: [Xen-users] Xen and PPC and ARM,
Mark Williamson <=
|
|
|
|
|